Booster pump

General information

The booster pump is a field of the machine Learning (Artificial intelligence connects). It is a principle which gathers many algorithms which are based on binary whole of classifieurs: the booster pump optimizes their performances.

The principle is resulting from the combination of classifieurs (also called assumptions). By successive iterations, the knowledge of a weak classifior - weak to classify - is added to the final classifior - strong to classify .

One calls learning weak an algorithm which provides weak classifieurs, able to recognize 2 classes at least as well as the chance would not do it (i.e. it is not mistaken more once on two on average). The provided classifior is balanced by the quality of his classification: the well it classifies, the more it will be important. The badly classified examples are boostés so that they have more importance with respect to the learning weak one with the next turn, so that it mitigates the lack.

One of the algorithms most used in booster pump is called AdaBoost, which means adaptive booster pump .

The booster pump is based on the Cadre CAP.

Principal methods related to the booster pump

  • Bootstraping
  • Bagging
  • Booster pump
  • AdaBoost
  • DiscreteAB
  • RealAB
  • GentleAB
  • WeightBoost
  • BrownBoost
  • LogitBoost
  • AsymBoost
  • KLBoost
  • FloatBoost
  • GloBoost

Other related methods

  • ADDEMUP

  • arcing: adaptive recombination off classifiers
  • bagging: bootstrap aggregation
  • bag-stacking: bagging more stacking
  • cascading
  • combination off classifiers
  • committees off networks
  • consensus theory
  • cragging: cross-country race aggregation (like k-fold cross-country race validation)
  • dagging: aggregation
  • dag-stacking disjoins-sample: dagging more stacking
  • divide and conquer classifiers
  • hagging: half-sample aggregation
  • mixture off expert
  • multiple to classify systems:
  • multi-training course and multi-level classifiers
  • OLC: optimal linear combination
  • pandemonium off reflective agents
  • sieving algorithms
  • stacking: feeding outputs off several models (and possibly the
original inputs) into has second-level model
  • voting

Bonds

  • http://www.grappa.univ-lille3.fr/~torre/guide.php?id=coursapprentissage
  • http://www.boosting.org/

Random links:Method of the secant | Battle of Al-buhayra | Almamy | Michel Fromentoux | SR 88 | Lamar,_Nébraska